Audiobox

Audiobox is a foundation research model from Meta that generates voices and sound effects from a combination of voice input and text prompts in natural language. The technology opens the door for users to create their own audio for a wide range of use cases, including speech synthesis and sound effects.

Audiobox is based on a family of special-purpose models, including Audiobox Speech and Audiobox Sound, all trained with a common self-supervised model called Audiobox SSL. You can try it out with interactive audio demos that let you experiment with what's possible. The website also includes tools like Audiobox Maker that lets you create your own audio stories by mixing and matching different elements.

Some of the things you can do with Audiobox include:

• Create Audio: Synthesize speech based on an audio sample or based on text prompts.
• Edit Audio: Remove background noise from an audio recording or replace parts with new audio.

Audiobox is intended for creative and experimental uses. For example, you can create voices with new styles or modify the style of an audio sample based on text prompts.

Audiobox is a research model, and the demos are for noncommercial use only. The service also isn't available to residents of Illinois and Texas. Before you try the demos, you'll have to agree to the Audiobox Demo Supplemental Terms of Service and Acceptable Use Policy.

Audiobox is powerful for generating audio, but as with all AI systems, it can produce incorrect results. Also, user data and output is collected and processed to improve the models. For the curious, Audiobox details the gory details on its blog post and research paper.

You can try the demos and explore the possibilities of this new audio generation technology on the Audiobox website.

Published on June 9, 2024